- Email: [email protected]
Emission-based attenuation correction of myocardial perfusion studies
Emission-based attenuation correction of myocardial perfusion studies M a r k T. M a d s e n , PhD, a P e t e r T. K i r c h n e r , M D , a M a l e a h G r o v e r - M c K a y , M D ) R e g a i A k t a y , M D , b J a m e s S. S e a b o l d , M D , a K a r i m Rezai, M D , ~ a n d G r e g Kelly, P A ~ Background. Nonuniform attenuation in the thorax can generate artifacts in single-photon emission computed tomographic myocardial perfusion studies that mimic coronary artery disease. In this article we present both phantom and simulation data, as well as clinical data, in support of an emission-based method that provides reliable correction for attenuation effects without the need for a transmission measurement. Methods and Results. The attenuation map is derived from the measured distribution of 99mTc-labeled macroaggregated albumin in the lungs and a radioactive binder wrapped about the thorax. This information is acquired as part of a dual-isotope acquisition during the rest 2°~TI study. Segmentation is used to define the interiors of lung and body compartments, which are assigned a single attenuation coefficient for each of the two tissue types. The appropriateness of this approach was investigated by examining the measured attenuation coefficients in a group of 80 individuals (40 male, 40 female) from positron emission tomographic transmission studies. The correction technique was evaluated with computer simulations, a physical phantom, and clinical data acquired from 20 patients. Analysis of the positron emission tomographic data found a small SD in the mean attenuation coefficients for the body ( < 5 % ) and lungs (<15%). The application of emission-based attenuation-correction technique produced a substantial reduction in the magnitude of the attenuation artifact in images obtained from both the phantom and the simulation studies. The emission-based attenuation-correction technique was easily applied to myocardial perfusion studies, where it had a significant effect, resulting in changes in interpretation for nine of 20 patients. Conclusions. The results of this study provide strong support for the concept that an attenuation map can be generated with fixed attenuation values in place of those that are directly measured. Thus the emission-based attenuation-correction technique can be considered an inexpensive alternative to transmission-based correction methods. Because the emission-based correction technique does not require any additional hardware, it has the major advantage of being applicable to all single-photon emission computed tomographic systems. (J Nucl Cardiol 1997;4:477-86.) Key Words: SPECT • attenuation correction • myocardial perfusion
Radionuclide myocardial perfusion single-photon emission computed tomographic (SPECT) studies are routinely used to evaluate patients for coronary artery disease. However, a significant fraction of these myocardial SPECT studies are difficult to interpret because of the nonuniform attenuation of the emitted photons by thoracic tissues. 1,2 Conventional methods of attenuation From the aUniversity of Iowa, Department of Radiology, Iowa City, and the ~Medical College of Wisconsin, Department of Radiology, Milwaukee. Received for pnblicat!on Feb. 4, 1997; accepted May 20, 1997. Reprint requests: Mark T. Madsen, PhD, University of Iowa, Department of Radiology, 200 Hawkins Dr., Iowa City, IA 52242. Copyright © 1997 by American Society of Nuclear Cardiology. 1071-3581/97/$5.00 + 0 43/1/834@3
correction are inadequate, because the thorax does not conform to the assumptions implicit in the standard correction algorithms; that is, it has neither an elliptic contour nor a uniform tissue density? An accurate attenuation correction is facilitated if the three-dimensional distribution of tissue attenuation coefficients is known. Many investigators have reported on techniques designed to obtain this information from gamma camera computed tomographic scans in which an external radionuclide is used as a transmission source and the gamma camera is used as the detector.4-~5 Such techniques have been successfully implemented and are now commercially available. However, there are disadvantages with this approach. Additional hardware is required to mount the external source for the transmis477
478
Madsen et al. Attenuation correction of myocardial perfusion studies
sion scans. New SPECT systems equipped with such hardware will be more expensive, and retrofits of transmission hardware for older SPECT systems are not universally available. W e investigated a different approach to the problem, one that is based entirely on emission data. This approach is based on the premise that for practical purposes the thorax consists of two tissue types, lung and unitdensity soft tissue, and that the attenuation coefficients for these tissues vary little both within and among individuals. If this premise is true, then the only information necessary for the creation of an attenuation map is definition of the boundaries of the lung and the subject's body contour. The attenuation map can be generated by assigning the appropriate attenuation coefficients for the g a m m a rays used in the study to the spaces defined by these boundaries. Other investigators have used these assumptions with transmission-generated attenuation maps to reduce the high levels of image noise that are often associated with the transmission method. 9 In a previou s study, we investigated the feasibility of an emission-based approach with 17.5 M B q (0.5 mCi) 99mTc-labeled macroaggregated albumin ( M A A ) to define the lung boundaries and a radioactive flexible body binder to accurately define the outer contour of the chest. 16 The results of that study demonstrated that this approach could be used to define the lung and body boundaries. It also demonstrated that the presence of extra radioactivity in the field of view did not substantially alter the reconstructed activity distribution in the heart, nor did it change the clinical interpretation. Since that time, several other groups have investigated similar approaches. 17-19 In this article we describe the validation of the emission-based attenuation correction. The variability of attenuation coefficients in the thorax was assessed by examining the transmission studies obtained from patients undergoing positron emission tomographic (PET) studies. The effect of substituting fixed attenuation values for the lung and unit-density tissues was evaluated from computer simulations of myocardial perfusion studies. SPECT studies were performed on an anthropomorphic Phantom to test the correction algorithm in a realistic imaging setting. Finally, the emission-based attenuation-correction technique was used on a group of 20 patients to assess its potential impact on interpretation of clinical studies. METHODS Tissue Attenuation Coefficients. Because the emission-based correction method is based on the assumption that the variability of tissue attenuation coefficients among individ-
Journal of Nuclear Cardiology November/December 1997 uals is small, the tissue attenuation coefficients of 80 patients (40 male and 40 female) who had PET transmission studies of the thorax were surveyed. The transmission data were collected in coincidence mode by means of a rotating pin source of 68Ge within a PET image. Gamma rays with an energy greater than 60 keV interact with soft tissue primarily through Compton scattering, which depends on density. Therefore the attenuation coefficients in the lungs and other soft body tissues for the x-rays of 2°1T1 or the gamma rays of 99mTc should be related to the measured values at 511 keV by a linear scale factor. Fifteen transverse slices from the reconstructed PET transmission data were selected to cover a 10 cm axial section centered on the heart of each subject. Each transverse slice was segmented into body and lung compartments by means of thresholds. Unitdensity soft-tissue regions were defined by pixel attenuation coefficients greater than 0.075/cm, whereas lung regions were defined by pixel attenuation coefficients less than 0.05/cm. The mean attenuation coefficients for unit-density soft tissue and the lungs were calculated for each slice and for the entire 10 cm section. Simulation Studies. The appropriateness of using fixed attenuation coefficients in the emission-based method was further investigated with a set of simulated SPECT myocardial perfusion studies. To have realistic spatial distributions of radioactivity and attenuation coefficients, the transmission and emission data sets used in the simulation were obtained from 10 clinical PET F-18 FDG patient studies performed (five male and five female patients). The reconstructed transmission data sets were scaled to match the expected attenuation coefficients at 140 keV. These registered data sets were used to generate 72 attenuated projection views that included the entire heart. The simulation used exponential attenuation but did not include the effects of scatter, spatial resolution, or image noise. The simulated projection views were used to create the following reconstructed image sets: (1) reconstructed heart images with no correction for attenuation, (2) reconstructed heart images with the true attenuation correction calculated from the transmission images used in the simulation, and (3) reconstructed heart images that were corrected for attenuation with the assumed attenuation coefficients for the unit-density soft-tissue and lung regions. Attenuation correction was performed with the method described here. Bull's-eye plots were generated for each of the data sets with commercially available routines (CEqual program; Siemens Medical Systems, Hoffman Estates, Ill.). Phantom Studies. Studies were acquired by means of an anthropomorphic thorax phantom (Data Spectrum Corporation, Hillsborough, N.C.) to validate the method. The thorax phantom consisted of an elliptic chamber (major axis 31.5 cm, minor axis 23.5 cm, length 18 cm) with a separate compartment for each lung and a 3.7 cm diameter nylon rod to simulate the spine. The lung compartments were filled with polystyrene foam beads and water to yield a mean density of 0.36 gm/cm 3, as measured by a transmission scan with a 6SGe pin source on a PET image. The mean density of the nylon rod was 1.88 gm/cm 3. The thorax phantom also contained a removable heart phantom with separate wall and chamber compartments. The myocardial wall compartment was filled with 29.6 MBq (800
Journal of Nuclear Cardiology Volume 4, Number 6;477-86 ixCi) 2roT1. An additional 74 MBq (2 mCi) 2°1T1was added to the elliptic chamber. A total of 10 MBq (270 IxCi) 99mTc was added to the lung chambers. A 1000 ml bag of saline solution was attached to the outside of the thorax phantom to simulate breast or diaphragmatic attenuation. The radioactive binder was wrapped around the phantom and the bag of saline solution. SPECT studies of the phantom were acquired by the dualisotope acquisition method described here. The measurements were repeated with a 2 × 1 cm inferior wall defect inserted into the heart phantom. All studies were processed and corrected for attenuation by means of the procedure described for the patient studies. The reconstructed images of the heart phantom were oriented into the standard short-axis views and were also displayed as bull's-eye plots. Patient Studies. Subjects participating in this investigation were recruited from patients referred to the clinic for myocardial perfusion studies. Informed consent was obtained from each of the 20 subjects (10 male and 10 female). At our institution myocardial perfusion is evaluated at rest with a 2°iT1 SPECT study and during subsequent stress by a 99mTc-labeled sestamibi SPECT study. The patient was injected at rest with 111 MBq (3 mCi) 2°1T1. As part of this investigation, the patient was also injected with 17.5 MBq (0.5 mCi) 99mTclabeled MAA to define the lung boundaries. Furthermore, before positioning for the SPECT acquisition a radioactive binder was wrapped about the patient's chest to define the body contour. The binder was prepared in advance by uniformly wetting a flexible abdominal binder (50 cm long × 20 cm wide) with a solution consisting of 17.5 MBq (500 txCi) 99mTcmixed into 40 ml isopropyl alcohol. The radioactive binder was placed in a fume hood until completely dry (approximately 10 minutes). The resting 2°aT1 data were acquired as a dual-isotope SPECT study on a three-detector SPECT system with energy windows centered on the Hg x-rays and the 140 keV photopeak of 99mTc. High-resolution collimators were used to acquire a total of 72 projection views from each energy window (24 stops at 5-degree angular increments) with a dwell time of 45 seconds. The image matrix was t28 × 128, with a pixel size of 3.56 mm. This provided sufficient data sampling and allowed full use of the camera field of view, minimizing the possibility of truncating portions of the body. When the first SPECT study was completed, the binder was removed and the patient was given 925 MBq (25 mCi) 99mTc-labeled sestamibi after an exercise or coronary vasodilator stress protocol. Immediately before the beginning of the second SPECT acquisition, the radioactive binder was replaced. The 99mTc-labeled sestamibi SPECT study was acquired as described previously, but with a single energy window centered on the 99mTc photopeak. The uncorrected and attenuation-corrected patient studies were reconstructed and oriented into the standard vertical long-axis, horizontal long-axis, and short axis views. For each subject, both uncorrected and attenuation-corrected SPECT images were generated and recorded on separate films, each film displaying the same format of rest and stress views. These films were interpreted in random order by four experienced nuclear medicine physicians who were blinded to subject identity but had knowledge of each subject's medical history. The perfusion status of each of the standard 20 wall segments
Madsen et al. Attenuationcorrection of myocardial perfusion studies
479
was ranked with a 0 to 4 ROC measure (0, normal; 1, probably normal; 2, equivocal; 3, probably abnormal; and 4, definitely abnormal) by the consensus of this group. Attenuation Correction. Attenuation correction requires a map of the attenuation coefficients coregistered with the myocardial perfusion data. This can be directly obtained for the dual-isotope 2°IT1 study as described here. Because of the difficulty in separating the lungs from the heart and liver in the 99mTc-labeled sestamibi studies, the attenuation map generated from the 2°iT1 study was also used for the correction of those studies. Thus the method of correcting the acquired myocardial perfusion studies for attenuation had six steps: (1) alignment of the 2°1T1 and 99mTc-labeled sestamibi projection images, (2) uncorrected filtered back-projection reconstruction of the 2°IT1 and 99mTc-labeledsestamibi heart studies and the 99mTc-labeled MAA lung data, (3) axial compression of the reconstructed images, (4) segmentation of the 99mTclung data and generation of the attenuation map, (5) calculation of the attenuationcorrection factors, and (6) application of the correction factors to the projection views and filtered back-projection reconstruction. Initial alignment of the heart projection views is accomplished by centering rectangular regions on the heart in projection views from both the 2°1T1 and 99mTc-labeled sestamibi studies. Because the patient's location within the imaging system is likely to change with each procedure, the coordinates from these regions are used to axially align the two SPECT studies. These coordinates also define the appropriate portion of the 99mTc-labeledMAA lung images corresponding to the level of the heart. Further alignment of the 99mTc-labeled sestamibi and lung data is performed on the reconstructed slices. These three data sets (2°1T1, 99mTc-labeled sestamibi, and 99mTclabeled MAA) are reconstructed with all 72 projection views by filtered back projection (Butterworth filter, 0.49/cm cutoff, order 5), and the transverse sections are saved. An axial compression is performed in the next step, Sequential groups of three contiguous axial slices are added together for each of the three data sets (2roT1, lung, and 99mTc-labeled sestamibi). This is done to decrease the computation time needed to obtain the correction factors for the projection images. Variations in attenuation-correction factors over an axial distance of 10.68 mm (3 pixels) are unlikely to have a big effect on the correction factors because the spatial resolution is on the order of 12 mm full width at half maximum. The attenuation map used in the correction algorithm is derived from segmenting the reconstructed 99mTc-labeled MAA data obtained from the dual-isotope study (Figure 1, A). The body contour is found first through a radial search for local maxima at 32 angles around the image of the body outline resulting from the radioactive binder. These points are fitted with a Fourier series to eliminate any bad points, and a connected outline of the body is formed. This process is repeated on subsequent slices. To prevent the search from straying from the vicinity of the body contour, this outline is used to define the search limits in the next transaxial slice. The new search starts 4 pixels radially inward from the previous outline and proceeds radially outward until a local maximum is found within the limit of 15 pixels beyond the starting point. The interior of the contour is filled with a value corre-
480
Madsenet al. Attenuationcorrection of myocardialperfusion studies
Journal of Nuclear Cardiology November/December 1997
Figure 1. Emission-based attenuation-correction method. A, Transverse section from the 99mTc window shows location of the lungs (arrow) and body contour (intensified outline). B, This information is segmented to form an attenuation map corresponding to that slice. I, Lung regions; 3, unit-density soft-tissue; 4, table. C, Morozumi method is used to calculate attenuation-correction factors. True and attenuated projections are calculated from uncorrected heart data. D, Corresponding attenuation map. E, Ratio of true and attenuated projections calculated from C and D becomes the correction factor for the acquired projection views. F, Shows how all the information is organized in the patient SPECT study. Region a holds the 99mTcprojection views. Region b contains the raw heart projection views. Region c contains the attenuation-corrected heart projection views, and region d has the correction factors for the view. Thus region c represents the product of the data in regions b and d. sponding to the attenuation coefficient for unit-density soft tissue. The boundaries of the lungs are determined by applying a count threshold to the 99mTc-labeledMAA images. All pixels within the lung that are greater than 20% of the maximum lung count value are set to the mean attenuation coefficient for lung tissue determined from the PET transmission scans and scaled to the appropriate photon energy. The 20% threshold is low enough to reliably yield lung fields corresponding to the MAA distribution but is also high enough to eliminate spurious counts from nonlung portions of the image. The lung information is combined with the unit-density soft-tissue map to yield the attenuation map. To this map we also add the attenuation coefficients for the table that supports the.patient during the scan (Figure 1, B). Because the attenuation map is derived from the dualisotope study, it is automatically aligned with the 2°1T1 heart images. Even though the projection views of the 2roT1 and 99mTc-labeled sestamibi are aligned axially at the beginning of the attenuation-correction process, an additional step is needed to verify that the attenuation map is aligned with the reconstructed 99mTc-labeledsestamibi images. This is done manually by translating a body contour region from the attenuation map onto the corresponding 99mTc-labeled sestamibi slice until the two appear registered. With the three-detector SPECT system, simple translation operations appear sufficient to achieve registration to within one pixel because the patient's position on the imaging table is somewhat constrained by the proximity of the detectors. Verification of the alignment in the orthogonal
planes may be necessary in cases Where the patient's orientation on the table is significantly different between the two studies, or in cases where the table is cantilevered and the degree of bending is different between studies. The attenuation-correctionfactors for the projection views are determined from the method originally described by Morozumi et al. 2° Taking the reconstructed emission data, E(x,y), as an estimate of the true internal distribution, the ratio of the true and attenuated projections is calculated at each acquired projection angle as illustrated in Figure 1 (C and D). The true projections at angle 0, PT(r,O), are the sum of the counts along each projection ray, NE(x,y)As, whereas the attenuated projections, PA(r,(~), require both the emission data and the attenuation map: ~E(x,y)As exp[-2u(x,y)As]. The correction factors obtained as the ratio of these two values are stored as a sinogram for each transaxial slice (Figure 1, E). When this process has been completed for both the 2°1T1and 99mTc data, the correction factors are interpolated in the axial direction to 3.56 mm/pixel and are multiplied into the original projection images (Figure 1, F). The corrected projections are then reconstructed using filtered back projection (Butterworth filter, 0.35 cutoff, order 5).
RESULTS Attenuation Coefficients. The mean attenuation coefficients measured in the 80 subjects are given in
Journal of Nuclear Cardiology Volume 4, Number 6;477-86
Madsen et al. Attenuation correction of myocardial perfusion studies
481
Table I . M e a n body* a n d lung a t t e n u a t i o n coefficients from PET t r a n s m i s s i o n d a t a Age ( y r )
Height (cm)
Weight (kg)
Surface area (mz)
rbody
57 _+ 12.5 83 22 0.09
rlung
--0.21
170 _+ 8.1 183 147 0.23 0.04
78.3 -- 16.4 121 42 --0.08 O. 19
56 + 3.3 74 34
175 _+ 5.3 183 165
58 -+ 14.9 83 22
165 -+ 7.4 183 147
All subjects (N = 80) Mean Max Min
Male subjects (n = 40) Mean Max Min Female subjects (n = 40) Mean Max Min
Body (cm -1)
Lung (cm -I)
1.90 + 0.21 2.28 1.43 0.03 O. 18
0.093 -+ 0.003 0.098 0.085
0.035 + 0.004 0.041 0.027
82.6 _+ 16.4 109 42
1.98 -+ 0.17 2.23 1.52
0.095 -+ 0.001 0.098 0.093
0.036 + 0.004 0.041 0.029
74.2 _+ 16.9 121 47
1.82 ,+ 0.21 2.28 1.43
0.090 -+ 0.003 0.093 0.085
0.034 + 0.003 0.040 0.027
rbody and rlu.g are the correlation coefficients found between the respective attenuation coefficients and the physical descriptors. *Body refers to unit-density soft tissues.
Table 1. The mean attenuation coefficient for unitdensity tissue at 511 keV was found to be 0.092 -+ 0.003/cm and the mean lung attenuation coefficient was found to be 0.035 -+ 0.004/cm. These values are in good agreement with the mean attenuation coefficients for these tissues reported in the literature 2l and are consistent with the assumption that tissue attenuation coefficients do not vary widely among individuals. No correlation existed between the attenuation coefficient values and subject age, weight, height, or body surface area for either the body or lungs. The mean body attenuation coefficient was slightly higher in male (0.095/cm) than female (0.090) subjects. This difference, although small, was statistically significant (p < 0.001). Simulations. The b u l l ' s - e y e plots from the 10 simulated studies are displayed in Figure 2. The first column (A) shows the results when the actual attenuation map was used for the attenuation correction. The second column (B) shows the results that were obtained when uniform mean attenuation coefficients were substituted in the lung and body regions and the Morozumi attenuation correction was applied. The third column ((7) shows the results obtained when no attenuation correction was applied. Comparisons of the b u l l ' s - e y e plots from the these simulated studies reveals a substantial difference between the corrected and uncorrected images and subtle differences between the results from a single iteration and the true correction. Without attenuation correction, regional counts were often more than 25%
lower than corresponding areas of the attenuation-corrected studies. On the other hand, the m a x i m u m regional count difference between bull' s-eye images created with an accurate attenuation correction and the bull's-eye images created with the proposed correction technique was less than 10%. This result is consistent with the phantom result and supports the idea that a single iteration is sufficient to provide a useful correction. Thorax Phantom. The results of the phantom studies indicate that the proposed method does correct attenuation artifacts when the assumptions of uniform tissue attenuation within the lung and body compartments are met. There was a 50% reduction in the count density of the inferior wall when the studies were reconstructed without attenuation correction. When attenuation correction was applied, the attenuation artifact was eliminated and the mean count density variation was less than 10% across the entire heart. This is illustrated in both Figures 3 and 4, which show the standard views and the bull' s-eye display for representative studies both with and without attenuation correction. Figure 3 shows the performance of the technique on a simulated normal heart and Figure 4 shows the performance with a simulated inferior wall defect in the inferior wall of the heart phantom. The attenuation-correction technique removes the artifact from the inferior wall and thereby improves the delineation of the simulated myocardial defect. Adequate attenuation correction of the phantom studies was achieved with one iteration of the Morozumi
482
Madsenet al. Attenuation correction of myocardial perfusion studies
A
B
C
I
|
Journal of Nuclear Cardiology November/December 1997
+%
B
C
m
!
3
4
9
5
l0
Figure 2. Comparison of bull's-eye plots derived from simulated heart studies. These images demonstrate that absolute accuracy of the attenuation map is not required for substantial reduction of the attenuation firtifact. An enhanced gray scale was used to emphasize the subtle differences in the displayed images. A, Accurate attenuation correction. B, Attenuation correction with mean attenuation coefficients for lung and body regions and one iteration of Morozumi method. C, No attenuation correction.
Figure 3. Comparison of uncorrected and attenuation-corrected phantom heart studies. A, Images on the left show the uncorrected bull's-eye image and representative images from the three orthogonal views. A 1000 ml bag of saline solution was used to create the inferior wall attenuation artifact. B, Images on the right show the corresponding views after attenuation correction was applied. algorithm. Figure 5 shows the b u l l ' s - e y e display for an uncorrected study, a first-order correction, and a secondorder correction. A single iteration corrects the regional count density to better than 90% of its true value. Although there is some further improvement associated
with the application of a second iteration, it is minimal and does not substantially alter the information obtained from the first-order correction. This result suggests that a single iteration may be sufficient for the correction of clinical studies.
Journal of Nuclear Cardiology Volume 4, Number 6;477-86
Madsen et al. Attenuation correction of myocardial perfusion studies
483
Figure 4. Comparison of uncorrected and attenuation-corrected phantom heart studies with a true inferior wall defect. A, Images on the left show the uncorrected bull's-eye image and representative images from the three orthogonal views. B, Images on the right show the corresponding views after attenuation correction was applied. Note that although the attenuation artifact is corrected, the true defect is plainly evident.
Figure 5. Comparison of first- and second-order iterations of attenuation correction. A, Uncorrected bull' s-eye image of a phantom heart. B, Bull' s-eye image of the same study after one iteration of Morozumi attenuation correction. C, Bull's-eye image after two iterations of Morozumi attenuation correction. Although there is additional subtle improvement with the second iteration, attenuation artifact is substantially eliminated with the first-order correction. Patients. Results of the patient studies are summarized in Tables 2 and 3. The application of attenuation correction changed the ROC scores in 39 of 400 segments (9.75%). The change in ROC confidence scores was two levels or greater in only 4.25% of the segments
(17/400), but this resulted in a change in the interpretation for eight of the 20 patients who participated in the study. Table 3 summarizes how the scan interpretation changed for these 17 segments. The attenuation correction tended to change the interpretation most often from
484
Madsenet al. Attenuation correction of myocardialperfusion studies
Journal of Nuclear Cardiology November/December1997 Table 3. S e g m e n t interpretation (ROC) changes
after attenuation correction Changes in s e g m e n t interpretation
Equivocal to normal (2 --~ 0) Equivocal to abnormal (2 --~ 4) Abnormal to normal (3 o r 4 ~ 0 o r 1) Normal to abnormal (0or i --~3or4) Total Figure 6. Comparison of (A) uncorrected and (B) attenuationcorrected bull's-eye images from five of the 20 patient studies. The method increases counts in regions susceptible to attenuation artifact (e.g., inferior wall) without obscuring true defects or introducing other artifacts. Table 2. Difference in ROC level b e t w e e n corrected
and uncorrected studies Segments involved (n = 400) Uncorrected vs attenuation correction
0 1 step 2 steps 3 steps
No.
%
361 22 9 8
91.25 5.5 2.2 2
normal to abnormal. For only five of the 17 changed segments did the visual assessment change from equivocal or abnormal to normal with the application of attenuation correction, whereas for 12 segments the interpretation changed from equivocal or normal to abnormal. This result suggests that the physician readers were used to seeing attenuation artifacts and compensated for them in their interpretation. The attenuationcorrected studies show a significant increase in the count density in regions susceptible to attenuation artifacts, without obscuring the focal abnormalities seen in the uncorrected images. This is illustrated in Figure 6, which shows corrected and uncorrected bull' s-eye plots for five of the 20 patient studies. DISCUSSION
The emission-based attenuation-correction technique has several obvious advantages. It is relatively inexpensive and it does not require any additional hard-
Segments involved (n = 400)
2 1 3 11 17
ware. Thus it can be used with any SPECT system, whereas the commercially available transmission systems require the purchase of a new SPECT system. The additional radiation dose is relatively small. The lung dose from the 99mTc-labeled MAA is about 1.1 mGy (110 mrad), and the dose to the skin from the radioactive binder is less than 0.05 mGy (5 mrad). The radioactive binder is similar to the approach first proposed by Gullberg et al.,22 and it proved to be a useful way to determine the body contour. Because it is made of an elastic material, it form fits to the body. Wallis et al.ls speculated that such a technique might have trouble with concave geometries, such as when the patient has large breasts. We have not noted any problem in our limited sample of subjects (10 women). When on occasion there is an air gap that occurs between the breasts, the binder is taped to the skin. It should be emphasized that the binder should be thoroughly dry before Use and handled with gloves because it is possible for small amounts of radioactivity to contaminate hands. To date, none of our patient studies have resulted in any significant contamination of the subject's clothes from the binder. With the emission-based attenuation-correction technique, a single attenuation coefficient is assigned for the lung region and another is assigned to the unitdensity soft tissues. Although there is a potential for artifact generation if the assumed attenuation coefficients do not reflect reality, the results in Table 1 show that the measured attenuation coefficients in the body were uniform among individuals, with SD less than 5%. Attenuation coefficients for different lung regions were found to be more variable, with SD approximately 15% among individuals. However, several studies have shown that absolute accuracy of the lung attenuation coefficients is not required, and that deviations on this order do not result in significant artifacts in the attenuation-corrected images.16,18,19 This is consistent with the results that were found in the simulation studies, which showed only small
Journal of Nuclear Cardiology Volume 4, Number 6;477-86
differences between images generated with an exact attenuation map and the images generated with fixed attenuation coefficients for the two tissue types. The accurate definition of the boundaries between lung and other body tissues is an important concern. It is a basic assumption of the emission-based attenuationcorrection method that the 99mTc-labeled MAA distribution reflects body regions that truly have lung tissue density. This may not be the case in patients with certain types of pulmonary disease, such as emphysema, in which the perfused lung distribution is often much smaller than the lung air space. Thus it is possible that some patients with severe lung disease may have to be excluded from the application of this technique. The application of the emission-based attenuationcorrection technique increases the processing time over that required for routine SPECT studies because of the need for doing multiple reconstructions. In our implementation, two reconstructions are required for each SPECT acquisition. These consist of an initial reconstruction of the raw projection data and then the final reconstruction of the attenuation-corrected projection data. With a standard nuclear medicine computer system, the combined time to perform the entire attenuationcorrection routine on both the stress and rest studies has been uniformly less than 30 minutes. Our phantom and simulation results suggest that a single iteration of the Morozumi technique may be sufficient to provide a clinically reliable result. Although this is consistent with the results of other investigators who have studied the Morozumi technique, 23,24 further investigations need to be done in cases where corrections for scattered radiation and spatial resolution are likely to be important. It should be noted that the use of the Morozumi technique is not a requirement for the application of this correction. Once the attenuation map has been generated from the segmented 99mTc-labeled MAA lung images, it is possible to use any of the reconstruction routines that provide nonuniform attenuation correction. Among these algorithms, those based on maximum likelihood techniques are especially attractive because of their ability to model attenuation, Poisson noise, and other physical factors of the imaging systemY Our study sampled projection views from 360 degrees around the subject. Because a three-detector SPECT system was used, there was no additional time required to obtain these projections. Complete sampling over 360 degrees may be necessary for determining accurate lung boundaries, because attempts to use 180degree data resulted in distorted reconstructions of the 99mTc distribution. Further research is needed to evaluate whether it may be possible to alter the processing routines to compensate for this problem. The emission-based technique is easily applied to
Madsen et al. Attenuation correction of myocardial perfusion studies
485
myocardial perfusion studies done with 2°lTl because the dual-isotope acquisition provides a clean image of the lungs and body boundaries, which are registered with the myocardial perfusion data. When used in conjunction with a 99mTc-labeled myocardial tracer, the emissionbased technique may encounter serious difficulty in defining the boundaries of the lungs, especially in those regions where the heart and liver are adjacent to the lungs. Although it is possible that a "smart" segmentation program could be developed to reliably define all lung boundaries, an alternative would be to acquire a 99mTc-labeled MAA scan before the administration of the heart agent, such as in the approach proposed by Wallis et al. 18 After acquisition of the heart study, the two studies could be coregistered and the information in the 99mTc-labeled MAA study could then be used to generate the attenuation map. The obvious disadvantage of such an approach is the need for two separate SPECT acquisition studies, with the resultant potential for registration errors. Conclusions. The results of this study suggest that a clinically useful attenuation map can be generated with fixed attenuation values in place of those that are directly measured. Thus the emission-based attenuation-correction technique can be considered as an inexpensive alternative to transmission-based correction methods. Because the emission-based correction technique does not require any additional hardware, it can be used with all SPECT systems.
References 1. DePuey EG, Garcia EV. Optimal specificity of thallium-201 SPECT through recognition of imaging artifacts. J Nucl Med 1989;30:441-9. 2. Wackers FJ. Artifacts in planar and SPECT myocardial perfusion studies. Am J Card Imaging 1992;6:42-8. 3. Chang LT. A method for attenuation correction in radionuclide computed tomography. IEEE Trans Nucl Sci 1978;NS-25:638-43. 4. Malko JA, GullbergGT, Kowalsky WP, Van Heertum RL. A count-based algorithm for attenuation corrected volume determination using, data from an external flood source. J Nucl Med 1985;26:194-200. 5. Bailey DL, Hutton BF, Walter PJ. Improved SPECT using simultaneous emission and transmission tomography. J Nucl Med 1987;28:844-51. 7. Tsui BMW, Gullberg GT, Edgerton ER, Ballard JG, Perry JR, McCartney WH, et al. Correction of nonuniform attenuation in cardiac SPECT imaging. J Nucl Med 1989;30:497-507. 8. Manglos SH, Bassano DA, Thomas FD, Grossman ZD. Imaging of the human torso using cone beam transmission CT implemented on a rotating gamma camera. J Nucl Med 1992;33:150-66. 9. Galt JR, Cullom SJ, GarciaEV. SPECT quantification: a simplified method of attenuation and scatter correction for cardiac imaging. J Nucl Med 1992;33:2232-7. 10. Frey EC, Tsui BMW, Perry JR. Simultaneous acquisition of
486
11.
12.
13.
14.
15.
16.
17.
Madsen et al. Attenuation correction of myocardial perfusion studies emission and transmission data for improved T1-201 cardiac SPECT imaging using a Tc-99m transmission source. J Nucl Med 1992;33:2238-45. Tung CH, Gullberg GT, Zeng GL, Christian PE, Datz FL, Morgan HT. Non-uniform attenuation correction using simultaneous transmission and emission converging tomography. 1EEE Trans Nucl Sci 1992;39:134-43. Manglos SH, Bassano DA, Thomas FD. Cone-beam transmission computed tomography for nonuniform attenuation compensation of SPECT images. J Nucl Med 1993;32:1813-20. Jaszczak RJ, Gilland DR, Hanson MW, Jang S, Greer KC, Coleman RE. Fast transmission CT for determining attenuation maps using a collimated line source, rotatable air copper lead attenuators and fan beam collimation. J Nucl Med 1993;34:157786. Tan P, Bailey DL, Mielke SR, Eberl S, Fulton RR, Hutton BF. A scanning line source for simultaneous emission and transmission measurements in SPECT. J Nucl Med 1993;34:1752-60. Ficaro EP, Fessler JA, Ackermann RJ, Rogers WL, Corbett JR, Schwaiger M. Simultaneous transmission-emission thallium-201 cardiac SPECT: effect of attenuation correction on myocardial tracer distribution. J Nucl Med 1995;36:921-31. Madsen MT, Kirchner PT, Edlin J, Nathan M, Kahn D. An emission-based technique for obtaining attenuation correction data for myocardial SPECT studies. Nucl Med Commun 1993;14:68995. Turkington TG, Jaszczak RJ, Gilland, Jang S, Greer KL, Coleman RE. A non-uniform attenuation correction for quantitative thorax SPECT based on lung emission (abstract). J Nucl Med 1993;34: 190.
Journal of Nuclear Cardiology November/December 1997 18. Wallis JW, Miller TR, Koppel P. Attenuation correction in cardiac SPECT without a transmission measurement. J Nucl Med 1995; 36:506-12. 19. Pan TS, King MA, Devries DJ, Ljungberg M. Segmentation of the body and lungs from Compton scatter and photopeak window data in SPECT: a Monte-Carlo investigation. IEEE Trans Med Imaging 1996;15:13-24. 20. Morozumi T, NAkajima M, Ogawa K, Yuta S. Attenuation correction method using information of attenuation distribution for single photon CT. Med Imaging Technol 1984;2:20-8. 21. International Commission on Radiation Units and Measurements. Tissue substitutes in radiation dosimetry and measurement, report 44. Bethesda (MD): International Commission on Radiation Units and Measurements; 1989. 22. Gullberg GT, Malko JA, Eisner RL. Boundary determination methods for attenuation correction in single photon emission computed tomography. In: Esser PD, editor. Emission computed tomography: current trends. New York: The Society of Nuclear Medicine; 1983. p. 33-53. 23. Murase K, Itoh H, Mogami H, Ishine M, Kawamura M, Iio A, et al. A comparative study of attenuation correction algorithms in single photon emission computed tomography (SPECT). Eur J Nucl Med 1987;13:55-62. 24. Soares EJ, King MA, Gtick SJ, Villegas BJ. Calculation of lung heart ratios for single photon emission computed tomography. IEEE Trans Nucl Sci 1996;NS-43:1995-9. 25. King MA, Tsui BMW, Pan TS, Glick SJ, Soares EJ. Attenuation compensation for cardiac emission computed tomographic imaging: part 2. Attenuation compensation algorithms. J Nucl Cardiol 1996;3:55-63.
Radionuclide myocardial perfusion single-photon emission computed tomographic (SPECT) studies are routinely used to evaluate patients for coronary artery disease. However, a significant fraction of these myocardial SPECT studies are difficult to interpret because of the nonuniform attenuation of the emitted photons by thoracic tissues. 1,2 Conventional methods of attenuation From the aUniversity of Iowa, Department of Radiology, Iowa City, and the ~Medical College of Wisconsin, Department of Radiology, Milwaukee. Received for pnblicat!on Feb. 4, 1997; accepted May 20, 1997. Reprint requests: Mark T. Madsen, PhD, University of Iowa, Department of Radiology, 200 Hawkins Dr., Iowa City, IA 52242. Copyright © 1997 by American Society of Nuclear Cardiology. 1071-3581/97/$5.00 + 0 43/1/834@3
correction are inadequate, because the thorax does not conform to the assumptions implicit in the standard correction algorithms; that is, it has neither an elliptic contour nor a uniform tissue density? An accurate attenuation correction is facilitated if the three-dimensional distribution of tissue attenuation coefficients is known. Many investigators have reported on techniques designed to obtain this information from gamma camera computed tomographic scans in which an external radionuclide is used as a transmission source and the gamma camera is used as the detector.4-~5 Such techniques have been successfully implemented and are now commercially available. However, there are disadvantages with this approach. Additional hardware is required to mount the external source for the transmis477
478
Madsen et al. Attenuation correction of myocardial perfusion studies
sion scans. New SPECT systems equipped with such hardware will be more expensive, and retrofits of transmission hardware for older SPECT systems are not universally available. W e investigated a different approach to the problem, one that is based entirely on emission data. This approach is based on the premise that for practical purposes the thorax consists of two tissue types, lung and unitdensity soft tissue, and that the attenuation coefficients for these tissues vary little both within and among individuals. If this premise is true, then the only information necessary for the creation of an attenuation map is definition of the boundaries of the lung and the subject's body contour. The attenuation map can be generated by assigning the appropriate attenuation coefficients for the g a m m a rays used in the study to the spaces defined by these boundaries. Other investigators have used these assumptions with transmission-generated attenuation maps to reduce the high levels of image noise that are often associated with the transmission method. 9 In a previou s study, we investigated the feasibility of an emission-based approach with 17.5 M B q (0.5 mCi) 99mTc-labeled macroaggregated albumin ( M A A ) to define the lung boundaries and a radioactive flexible body binder to accurately define the outer contour of the chest. 16 The results of that study demonstrated that this approach could be used to define the lung and body boundaries. It also demonstrated that the presence of extra radioactivity in the field of view did not substantially alter the reconstructed activity distribution in the heart, nor did it change the clinical interpretation. Since that time, several other groups have investigated similar approaches. 17-19 In this article we describe the validation of the emission-based attenuation correction. The variability of attenuation coefficients in the thorax was assessed by examining the transmission studies obtained from patients undergoing positron emission tomographic (PET) studies. The effect of substituting fixed attenuation values for the lung and unit-density tissues was evaluated from computer simulations of myocardial perfusion studies. SPECT studies were performed on an anthropomorphic Phantom to test the correction algorithm in a realistic imaging setting. Finally, the emission-based attenuation-correction technique was used on a group of 20 patients to assess its potential impact on interpretation of clinical studies. METHODS Tissue Attenuation Coefficients. Because the emission-based correction method is based on the assumption that the variability of tissue attenuation coefficients among individ-
Journal of Nuclear Cardiology November/December 1997 uals is small, the tissue attenuation coefficients of 80 patients (40 male and 40 female) who had PET transmission studies of the thorax were surveyed. The transmission data were collected in coincidence mode by means of a rotating pin source of 68Ge within a PET image. Gamma rays with an energy greater than 60 keV interact with soft tissue primarily through Compton scattering, which depends on density. Therefore the attenuation coefficients in the lungs and other soft body tissues for the x-rays of 2°1T1 or the gamma rays of 99mTc should be related to the measured values at 511 keV by a linear scale factor. Fifteen transverse slices from the reconstructed PET transmission data were selected to cover a 10 cm axial section centered on the heart of each subject. Each transverse slice was segmented into body and lung compartments by means of thresholds. Unitdensity soft-tissue regions were defined by pixel attenuation coefficients greater than 0.075/cm, whereas lung regions were defined by pixel attenuation coefficients less than 0.05/cm. The mean attenuation coefficients for unit-density soft tissue and the lungs were calculated for each slice and for the entire 10 cm section. Simulation Studies. The appropriateness of using fixed attenuation coefficients in the emission-based method was further investigated with a set of simulated SPECT myocardial perfusion studies. To have realistic spatial distributions of radioactivity and attenuation coefficients, the transmission and emission data sets used in the simulation were obtained from 10 clinical PET F-18 FDG patient studies performed (five male and five female patients). The reconstructed transmission data sets were scaled to match the expected attenuation coefficients at 140 keV. These registered data sets were used to generate 72 attenuated projection views that included the entire heart. The simulation used exponential attenuation but did not include the effects of scatter, spatial resolution, or image noise. The simulated projection views were used to create the following reconstructed image sets: (1) reconstructed heart images with no correction for attenuation, (2) reconstructed heart images with the true attenuation correction calculated from the transmission images used in the simulation, and (3) reconstructed heart images that were corrected for attenuation with the assumed attenuation coefficients for the unit-density soft-tissue and lung regions. Attenuation correction was performed with the method described here. Bull's-eye plots were generated for each of the data sets with commercially available routines (CEqual program; Siemens Medical Systems, Hoffman Estates, Ill.). Phantom Studies. Studies were acquired by means of an anthropomorphic thorax phantom (Data Spectrum Corporation, Hillsborough, N.C.) to validate the method. The thorax phantom consisted of an elliptic chamber (major axis 31.5 cm, minor axis 23.5 cm, length 18 cm) with a separate compartment for each lung and a 3.7 cm diameter nylon rod to simulate the spine. The lung compartments were filled with polystyrene foam beads and water to yield a mean density of 0.36 gm/cm 3, as measured by a transmission scan with a 6SGe pin source on a PET image. The mean density of the nylon rod was 1.88 gm/cm 3. The thorax phantom also contained a removable heart phantom with separate wall and chamber compartments. The myocardial wall compartment was filled with 29.6 MBq (800
Journal of Nuclear Cardiology Volume 4, Number 6;477-86 ixCi) 2roT1. An additional 74 MBq (2 mCi) 2°1T1was added to the elliptic chamber. A total of 10 MBq (270 IxCi) 99mTc was added to the lung chambers. A 1000 ml bag of saline solution was attached to the outside of the thorax phantom to simulate breast or diaphragmatic attenuation. The radioactive binder was wrapped around the phantom and the bag of saline solution. SPECT studies of the phantom were acquired by the dualisotope acquisition method described here. The measurements were repeated with a 2 × 1 cm inferior wall defect inserted into the heart phantom. All studies were processed and corrected for attenuation by means of the procedure described for the patient studies. The reconstructed images of the heart phantom were oriented into the standard short-axis views and were also displayed as bull's-eye plots. Patient Studies. Subjects participating in this investigation were recruited from patients referred to the clinic for myocardial perfusion studies. Informed consent was obtained from each of the 20 subjects (10 male and 10 female). At our institution myocardial perfusion is evaluated at rest with a 2°iT1 SPECT study and during subsequent stress by a 99mTc-labeled sestamibi SPECT study. The patient was injected at rest with 111 MBq (3 mCi) 2°1T1. As part of this investigation, the patient was also injected with 17.5 MBq (0.5 mCi) 99mTclabeled MAA to define the lung boundaries. Furthermore, before positioning for the SPECT acquisition a radioactive binder was wrapped about the patient's chest to define the body contour. The binder was prepared in advance by uniformly wetting a flexible abdominal binder (50 cm long × 20 cm wide) with a solution consisting of 17.5 MBq (500 txCi) 99mTcmixed into 40 ml isopropyl alcohol. The radioactive binder was placed in a fume hood until completely dry (approximately 10 minutes). The resting 2°aT1 data were acquired as a dual-isotope SPECT study on a three-detector SPECT system with energy windows centered on the Hg x-rays and the 140 keV photopeak of 99mTc. High-resolution collimators were used to acquire a total of 72 projection views from each energy window (24 stops at 5-degree angular increments) with a dwell time of 45 seconds. The image matrix was t28 × 128, with a pixel size of 3.56 mm. This provided sufficient data sampling and allowed full use of the camera field of view, minimizing the possibility of truncating portions of the body. When the first SPECT study was completed, the binder was removed and the patient was given 925 MBq (25 mCi) 99mTc-labeled sestamibi after an exercise or coronary vasodilator stress protocol. Immediately before the beginning of the second SPECT acquisition, the radioactive binder was replaced. The 99mTc-labeled sestamibi SPECT study was acquired as described previously, but with a single energy window centered on the 99mTc photopeak. The uncorrected and attenuation-corrected patient studies were reconstructed and oriented into the standard vertical long-axis, horizontal long-axis, and short axis views. For each subject, both uncorrected and attenuation-corrected SPECT images were generated and recorded on separate films, each film displaying the same format of rest and stress views. These films were interpreted in random order by four experienced nuclear medicine physicians who were blinded to subject identity but had knowledge of each subject's medical history. The perfusion status of each of the standard 20 wall segments
Madsen et al. Attenuationcorrection of myocardial perfusion studies
479
was ranked with a 0 to 4 ROC measure (0, normal; 1, probably normal; 2, equivocal; 3, probably abnormal; and 4, definitely abnormal) by the consensus of this group. Attenuation Correction. Attenuation correction requires a map of the attenuation coefficients coregistered with the myocardial perfusion data. This can be directly obtained for the dual-isotope 2°IT1 study as described here. Because of the difficulty in separating the lungs from the heart and liver in the 99mTc-labeled sestamibi studies, the attenuation map generated from the 2°iT1 study was also used for the correction of those studies. Thus the method of correcting the acquired myocardial perfusion studies for attenuation had six steps: (1) alignment of the 2°1T1 and 99mTc-labeled sestamibi projection images, (2) uncorrected filtered back-projection reconstruction of the 2°IT1 and 99mTc-labeledsestamibi heart studies and the 99mTc-labeled MAA lung data, (3) axial compression of the reconstructed images, (4) segmentation of the 99mTclung data and generation of the attenuation map, (5) calculation of the attenuationcorrection factors, and (6) application of the correction factors to the projection views and filtered back-projection reconstruction. Initial alignment of the heart projection views is accomplished by centering rectangular regions on the heart in projection views from both the 2°1T1 and 99mTc-labeled sestamibi studies. Because the patient's location within the imaging system is likely to change with each procedure, the coordinates from these regions are used to axially align the two SPECT studies. These coordinates also define the appropriate portion of the 99mTc-labeledMAA lung images corresponding to the level of the heart. Further alignment of the 99mTc-labeled sestamibi and lung data is performed on the reconstructed slices. These three data sets (2°1T1, 99mTc-labeled sestamibi, and 99mTclabeled MAA) are reconstructed with all 72 projection views by filtered back projection (Butterworth filter, 0.49/cm cutoff, order 5), and the transverse sections are saved. An axial compression is performed in the next step, Sequential groups of three contiguous axial slices are added together for each of the three data sets (2roT1, lung, and 99mTc-labeled sestamibi). This is done to decrease the computation time needed to obtain the correction factors for the projection images. Variations in attenuation-correction factors over an axial distance of 10.68 mm (3 pixels) are unlikely to have a big effect on the correction factors because the spatial resolution is on the order of 12 mm full width at half maximum. The attenuation map used in the correction algorithm is derived from segmenting the reconstructed 99mTc-labeled MAA data obtained from the dual-isotope study (Figure 1, A). The body contour is found first through a radial search for local maxima at 32 angles around the image of the body outline resulting from the radioactive binder. These points are fitted with a Fourier series to eliminate any bad points, and a connected outline of the body is formed. This process is repeated on subsequent slices. To prevent the search from straying from the vicinity of the body contour, this outline is used to define the search limits in the next transaxial slice. The new search starts 4 pixels radially inward from the previous outline and proceeds radially outward until a local maximum is found within the limit of 15 pixels beyond the starting point. The interior of the contour is filled with a value corre-
480
Madsenet al. Attenuationcorrection of myocardialperfusion studies
Journal of Nuclear Cardiology November/December 1997
Figure 1. Emission-based attenuation-correction method. A, Transverse section from the 99mTc window shows location of the lungs (arrow) and body contour (intensified outline). B, This information is segmented to form an attenuation map corresponding to that slice. I, Lung regions; 3, unit-density soft-tissue; 4, table. C, Morozumi method is used to calculate attenuation-correction factors. True and attenuated projections are calculated from uncorrected heart data. D, Corresponding attenuation map. E, Ratio of true and attenuated projections calculated from C and D becomes the correction factor for the acquired projection views. F, Shows how all the information is organized in the patient SPECT study. Region a holds the 99mTcprojection views. Region b contains the raw heart projection views. Region c contains the attenuation-corrected heart projection views, and region d has the correction factors for the view. Thus region c represents the product of the data in regions b and d. sponding to the attenuation coefficient for unit-density soft tissue. The boundaries of the lungs are determined by applying a count threshold to the 99mTc-labeledMAA images. All pixels within the lung that are greater than 20% of the maximum lung count value are set to the mean attenuation coefficient for lung tissue determined from the PET transmission scans and scaled to the appropriate photon energy. The 20% threshold is low enough to reliably yield lung fields corresponding to the MAA distribution but is also high enough to eliminate spurious counts from nonlung portions of the image. The lung information is combined with the unit-density soft-tissue map to yield the attenuation map. To this map we also add the attenuation coefficients for the table that supports the.patient during the scan (Figure 1, B). Because the attenuation map is derived from the dualisotope study, it is automatically aligned with the 2°1T1 heart images. Even though the projection views of the 2roT1 and 99mTc-labeled sestamibi are aligned axially at the beginning of the attenuation-correction process, an additional step is needed to verify that the attenuation map is aligned with the reconstructed 99mTc-labeledsestamibi images. This is done manually by translating a body contour region from the attenuation map onto the corresponding 99mTc-labeled sestamibi slice until the two appear registered. With the three-detector SPECT system, simple translation operations appear sufficient to achieve registration to within one pixel because the patient's position on the imaging table is somewhat constrained by the proximity of the detectors. Verification of the alignment in the orthogonal
planes may be necessary in cases Where the patient's orientation on the table is significantly different between the two studies, or in cases where the table is cantilevered and the degree of bending is different between studies. The attenuation-correctionfactors for the projection views are determined from the method originally described by Morozumi et al. 2° Taking the reconstructed emission data, E(x,y), as an estimate of the true internal distribution, the ratio of the true and attenuated projections is calculated at each acquired projection angle as illustrated in Figure 1 (C and D). The true projections at angle 0, PT(r,O), are the sum of the counts along each projection ray, NE(x,y)As, whereas the attenuated projections, PA(r,(~), require both the emission data and the attenuation map: ~E(x,y)As exp[-2u(x,y)As]. The correction factors obtained as the ratio of these two values are stored as a sinogram for each transaxial slice (Figure 1, E). When this process has been completed for both the 2°1T1and 99mTc data, the correction factors are interpolated in the axial direction to 3.56 mm/pixel and are multiplied into the original projection images (Figure 1, F). The corrected projections are then reconstructed using filtered back projection (Butterworth filter, 0.35 cutoff, order 5).
RESULTS Attenuation Coefficients. The mean attenuation coefficients measured in the 80 subjects are given in
Journal of Nuclear Cardiology Volume 4, Number 6;477-86
Madsen et al. Attenuation correction of myocardial perfusion studies
481
Table I . M e a n body* a n d lung a t t e n u a t i o n coefficients from PET t r a n s m i s s i o n d a t a Age ( y r )
Height (cm)
Weight (kg)
Surface area (mz)
rbody
57 _+ 12.5 83 22 0.09
rlung
--0.21
170 _+ 8.1 183 147 0.23 0.04
78.3 -- 16.4 121 42 --0.08 O. 19
56 + 3.3 74 34
175 _+ 5.3 183 165
58 -+ 14.9 83 22
165 -+ 7.4 183 147
All subjects (N = 80) Mean Max Min
Male subjects (n = 40) Mean Max Min Female subjects (n = 40) Mean Max Min
Body (cm -1)
Lung (cm -I)
1.90 + 0.21 2.28 1.43 0.03 O. 18
0.093 -+ 0.003 0.098 0.085
0.035 + 0.004 0.041 0.027
82.6 _+ 16.4 109 42
1.98 -+ 0.17 2.23 1.52
0.095 -+ 0.001 0.098 0.093
0.036 + 0.004 0.041 0.029
74.2 _+ 16.9 121 47
1.82 ,+ 0.21 2.28 1.43
0.090 -+ 0.003 0.093 0.085
0.034 + 0.003 0.040 0.027
rbody and rlu.g are the correlation coefficients found between the respective attenuation coefficients and the physical descriptors. *Body refers to unit-density soft tissues.
Table 1. The mean attenuation coefficient for unitdensity tissue at 511 keV was found to be 0.092 -+ 0.003/cm and the mean lung attenuation coefficient was found to be 0.035 -+ 0.004/cm. These values are in good agreement with the mean attenuation coefficients for these tissues reported in the literature 2l and are consistent with the assumption that tissue attenuation coefficients do not vary widely among individuals. No correlation existed between the attenuation coefficient values and subject age, weight, height, or body surface area for either the body or lungs. The mean body attenuation coefficient was slightly higher in male (0.095/cm) than female (0.090) subjects. This difference, although small, was statistically significant (p < 0.001). Simulations. The b u l l ' s - e y e plots from the 10 simulated studies are displayed in Figure 2. The first column (A) shows the results when the actual attenuation map was used for the attenuation correction. The second column (B) shows the results that were obtained when uniform mean attenuation coefficients were substituted in the lung and body regions and the Morozumi attenuation correction was applied. The third column ((7) shows the results obtained when no attenuation correction was applied. Comparisons of the b u l l ' s - e y e plots from the these simulated studies reveals a substantial difference between the corrected and uncorrected images and subtle differences between the results from a single iteration and the true correction. Without attenuation correction, regional counts were often more than 25%
lower than corresponding areas of the attenuation-corrected studies. On the other hand, the m a x i m u m regional count difference between bull' s-eye images created with an accurate attenuation correction and the bull's-eye images created with the proposed correction technique was less than 10%. This result is consistent with the phantom result and supports the idea that a single iteration is sufficient to provide a useful correction. Thorax Phantom. The results of the phantom studies indicate that the proposed method does correct attenuation artifacts when the assumptions of uniform tissue attenuation within the lung and body compartments are met. There was a 50% reduction in the count density of the inferior wall when the studies were reconstructed without attenuation correction. When attenuation correction was applied, the attenuation artifact was eliminated and the mean count density variation was less than 10% across the entire heart. This is illustrated in both Figures 3 and 4, which show the standard views and the bull' s-eye display for representative studies both with and without attenuation correction. Figure 3 shows the performance of the technique on a simulated normal heart and Figure 4 shows the performance with a simulated inferior wall defect in the inferior wall of the heart phantom. The attenuation-correction technique removes the artifact from the inferior wall and thereby improves the delineation of the simulated myocardial defect. Adequate attenuation correction of the phantom studies was achieved with one iteration of the Morozumi
482
Madsenet al. Attenuation correction of myocardial perfusion studies
A
B
C
I
|
Journal of Nuclear Cardiology November/December 1997
+%
B
C
m
!
3
4
9
5
l0
Figure 2. Comparison of bull's-eye plots derived from simulated heart studies. These images demonstrate that absolute accuracy of the attenuation map is not required for substantial reduction of the attenuation firtifact. An enhanced gray scale was used to emphasize the subtle differences in the displayed images. A, Accurate attenuation correction. B, Attenuation correction with mean attenuation coefficients for lung and body regions and one iteration of Morozumi method. C, No attenuation correction.
Figure 3. Comparison of uncorrected and attenuation-corrected phantom heart studies. A, Images on the left show the uncorrected bull's-eye image and representative images from the three orthogonal views. A 1000 ml bag of saline solution was used to create the inferior wall attenuation artifact. B, Images on the right show the corresponding views after attenuation correction was applied. algorithm. Figure 5 shows the b u l l ' s - e y e display for an uncorrected study, a first-order correction, and a secondorder correction. A single iteration corrects the regional count density to better than 90% of its true value. Although there is some further improvement associated
with the application of a second iteration, it is minimal and does not substantially alter the information obtained from the first-order correction. This result suggests that a single iteration may be sufficient for the correction of clinical studies.
Journal of Nuclear Cardiology Volume 4, Number 6;477-86
Madsen et al. Attenuation correction of myocardial perfusion studies
483
Figure 4. Comparison of uncorrected and attenuation-corrected phantom heart studies with a true inferior wall defect. A, Images on the left show the uncorrected bull's-eye image and representative images from the three orthogonal views. B, Images on the right show the corresponding views after attenuation correction was applied. Note that although the attenuation artifact is corrected, the true defect is plainly evident.
Figure 5. Comparison of first- and second-order iterations of attenuation correction. A, Uncorrected bull' s-eye image of a phantom heart. B, Bull' s-eye image of the same study after one iteration of Morozumi attenuation correction. C, Bull's-eye image after two iterations of Morozumi attenuation correction. Although there is additional subtle improvement with the second iteration, attenuation artifact is substantially eliminated with the first-order correction. Patients. Results of the patient studies are summarized in Tables 2 and 3. The application of attenuation correction changed the ROC scores in 39 of 400 segments (9.75%). The change in ROC confidence scores was two levels or greater in only 4.25% of the segments
(17/400), but this resulted in a change in the interpretation for eight of the 20 patients who participated in the study. Table 3 summarizes how the scan interpretation changed for these 17 segments. The attenuation correction tended to change the interpretation most often from
484
Madsenet al. Attenuation correction of myocardialperfusion studies
Journal of Nuclear Cardiology November/December1997 Table 3. S e g m e n t interpretation (ROC) changes
after attenuation correction Changes in s e g m e n t interpretation
Equivocal to normal (2 --~ 0) Equivocal to abnormal (2 --~ 4) Abnormal to normal (3 o r 4 ~ 0 o r 1) Normal to abnormal (0or i --~3or4) Total Figure 6. Comparison of (A) uncorrected and (B) attenuationcorrected bull's-eye images from five of the 20 patient studies. The method increases counts in regions susceptible to attenuation artifact (e.g., inferior wall) without obscuring true defects or introducing other artifacts. Table 2. Difference in ROC level b e t w e e n corrected
and uncorrected studies Segments involved (n = 400) Uncorrected vs attenuation correction
0 1 step 2 steps 3 steps
No.
%
361 22 9 8
91.25 5.5 2.2 2
normal to abnormal. For only five of the 17 changed segments did the visual assessment change from equivocal or abnormal to normal with the application of attenuation correction, whereas for 12 segments the interpretation changed from equivocal or normal to abnormal. This result suggests that the physician readers were used to seeing attenuation artifacts and compensated for them in their interpretation. The attenuationcorrected studies show a significant increase in the count density in regions susceptible to attenuation artifacts, without obscuring the focal abnormalities seen in the uncorrected images. This is illustrated in Figure 6, which shows corrected and uncorrected bull' s-eye plots for five of the 20 patient studies. DISCUSSION
The emission-based attenuation-correction technique has several obvious advantages. It is relatively inexpensive and it does not require any additional hard-
Segments involved (n = 400)
2 1 3 11 17
ware. Thus it can be used with any SPECT system, whereas the commercially available transmission systems require the purchase of a new SPECT system. The additional radiation dose is relatively small. The lung dose from the 99mTc-labeled MAA is about 1.1 mGy (110 mrad), and the dose to the skin from the radioactive binder is less than 0.05 mGy (5 mrad). The radioactive binder is similar to the approach first proposed by Gullberg et al.,22 and it proved to be a useful way to determine the body contour. Because it is made of an elastic material, it form fits to the body. Wallis et al.ls speculated that such a technique might have trouble with concave geometries, such as when the patient has large breasts. We have not noted any problem in our limited sample of subjects (10 women). When on occasion there is an air gap that occurs between the breasts, the binder is taped to the skin. It should be emphasized that the binder should be thoroughly dry before Use and handled with gloves because it is possible for small amounts of radioactivity to contaminate hands. To date, none of our patient studies have resulted in any significant contamination of the subject's clothes from the binder. With the emission-based attenuation-correction technique, a single attenuation coefficient is assigned for the lung region and another is assigned to the unitdensity soft tissues. Although there is a potential for artifact generation if the assumed attenuation coefficients do not reflect reality, the results in Table 1 show that the measured attenuation coefficients in the body were uniform among individuals, with SD less than 5%. Attenuation coefficients for different lung regions were found to be more variable, with SD approximately 15% among individuals. However, several studies have shown that absolute accuracy of the lung attenuation coefficients is not required, and that deviations on this order do not result in significant artifacts in the attenuation-corrected images.16,18,19 This is consistent with the results that were found in the simulation studies, which showed only small
Journal of Nuclear Cardiology Volume 4, Number 6;477-86
differences between images generated with an exact attenuation map and the images generated with fixed attenuation coefficients for the two tissue types. The accurate definition of the boundaries between lung and other body tissues is an important concern. It is a basic assumption of the emission-based attenuationcorrection method that the 99mTc-labeled MAA distribution reflects body regions that truly have lung tissue density. This may not be the case in patients with certain types of pulmonary disease, such as emphysema, in which the perfused lung distribution is often much smaller than the lung air space. Thus it is possible that some patients with severe lung disease may have to be excluded from the application of this technique. The application of the emission-based attenuationcorrection technique increases the processing time over that required for routine SPECT studies because of the need for doing multiple reconstructions. In our implementation, two reconstructions are required for each SPECT acquisition. These consist of an initial reconstruction of the raw projection data and then the final reconstruction of the attenuation-corrected projection data. With a standard nuclear medicine computer system, the combined time to perform the entire attenuationcorrection routine on both the stress and rest studies has been uniformly less than 30 minutes. Our phantom and simulation results suggest that a single iteration of the Morozumi technique may be sufficient to provide a clinically reliable result. Although this is consistent with the results of other investigators who have studied the Morozumi technique, 23,24 further investigations need to be done in cases where corrections for scattered radiation and spatial resolution are likely to be important. It should be noted that the use of the Morozumi technique is not a requirement for the application of this correction. Once the attenuation map has been generated from the segmented 99mTc-labeled MAA lung images, it is possible to use any of the reconstruction routines that provide nonuniform attenuation correction. Among these algorithms, those based on maximum likelihood techniques are especially attractive because of their ability to model attenuation, Poisson noise, and other physical factors of the imaging systemY Our study sampled projection views from 360 degrees around the subject. Because a three-detector SPECT system was used, there was no additional time required to obtain these projections. Complete sampling over 360 degrees may be necessary for determining accurate lung boundaries, because attempts to use 180degree data resulted in distorted reconstructions of the 99mTc distribution. Further research is needed to evaluate whether it may be possible to alter the processing routines to compensate for this problem. The emission-based technique is easily applied to
Madsen et al. Attenuation correction of myocardial perfusion studies
485
myocardial perfusion studies done with 2°lTl because the dual-isotope acquisition provides a clean image of the lungs and body boundaries, which are registered with the myocardial perfusion data. When used in conjunction with a 99mTc-labeled myocardial tracer, the emissionbased technique may encounter serious difficulty in defining the boundaries of the lungs, especially in those regions where the heart and liver are adjacent to the lungs. Although it is possible that a "smart" segmentation program could be developed to reliably define all lung boundaries, an alternative would be to acquire a 99mTc-labeled MAA scan before the administration of the heart agent, such as in the approach proposed by Wallis et al. 18 After acquisition of the heart study, the two studies could be coregistered and the information in the 99mTc-labeled MAA study could then be used to generate the attenuation map. The obvious disadvantage of such an approach is the need for two separate SPECT acquisition studies, with the resultant potential for registration errors. Conclusions. The results of this study suggest that a clinically useful attenuation map can be generated with fixed attenuation values in place of those that are directly measured. Thus the emission-based attenuation-correction technique can be considered as an inexpensive alternative to transmission-based correction methods. Because the emission-based correction technique does not require any additional hardware, it can be used with all SPECT systems.
References 1. DePuey EG, Garcia EV. Optimal specificity of thallium-201 SPECT through recognition of imaging artifacts. J Nucl Med 1989;30:441-9. 2. Wackers FJ. Artifacts in planar and SPECT myocardial perfusion studies. Am J Card Imaging 1992;6:42-8. 3. Chang LT. A method for attenuation correction in radionuclide computed tomography. IEEE Trans Nucl Sci 1978;NS-25:638-43. 4. Malko JA, GullbergGT, Kowalsky WP, Van Heertum RL. A count-based algorithm for attenuation corrected volume determination using, data from an external flood source. J Nucl Med 1985;26:194-200. 5. Bailey DL, Hutton BF, Walter PJ. Improved SPECT using simultaneous emission and transmission tomography. J Nucl Med 1987;28:844-51. 7. Tsui BMW, Gullberg GT, Edgerton ER, Ballard JG, Perry JR, McCartney WH, et al. Correction of nonuniform attenuation in cardiac SPECT imaging. J Nucl Med 1989;30:497-507. 8. Manglos SH, Bassano DA, Thomas FD, Grossman ZD. Imaging of the human torso using cone beam transmission CT implemented on a rotating gamma camera. J Nucl Med 1992;33:150-66. 9. Galt JR, Cullom SJ, GarciaEV. SPECT quantification: a simplified method of attenuation and scatter correction for cardiac imaging. J Nucl Med 1992;33:2232-7. 10. Frey EC, Tsui BMW, Perry JR. Simultaneous acquisition of
486
11.
12.
13.
14.
15.
16.
17.
Madsen et al. Attenuation correction of myocardial perfusion studies emission and transmission data for improved T1-201 cardiac SPECT imaging using a Tc-99m transmission source. J Nucl Med 1992;33:2238-45. Tung CH, Gullberg GT, Zeng GL, Christian PE, Datz FL, Morgan HT. Non-uniform attenuation correction using simultaneous transmission and emission converging tomography. 1EEE Trans Nucl Sci 1992;39:134-43. Manglos SH, Bassano DA, Thomas FD. Cone-beam transmission computed tomography for nonuniform attenuation compensation of SPECT images. J Nucl Med 1993;32:1813-20. Jaszczak RJ, Gilland DR, Hanson MW, Jang S, Greer KC, Coleman RE. Fast transmission CT for determining attenuation maps using a collimated line source, rotatable air copper lead attenuators and fan beam collimation. J Nucl Med 1993;34:157786. Tan P, Bailey DL, Mielke SR, Eberl S, Fulton RR, Hutton BF. A scanning line source for simultaneous emission and transmission measurements in SPECT. J Nucl Med 1993;34:1752-60. Ficaro EP, Fessler JA, Ackermann RJ, Rogers WL, Corbett JR, Schwaiger M. Simultaneous transmission-emission thallium-201 cardiac SPECT: effect of attenuation correction on myocardial tracer distribution. J Nucl Med 1995;36:921-31. Madsen MT, Kirchner PT, Edlin J, Nathan M, Kahn D. An emission-based technique for obtaining attenuation correction data for myocardial SPECT studies. Nucl Med Commun 1993;14:68995. Turkington TG, Jaszczak RJ, Gilland, Jang S, Greer KL, Coleman RE. A non-uniform attenuation correction for quantitative thorax SPECT based on lung emission (abstract). J Nucl Med 1993;34: 190.
Journal of Nuclear Cardiology November/December 1997 18. Wallis JW, Miller TR, Koppel P. Attenuation correction in cardiac SPECT without a transmission measurement. J Nucl Med 1995; 36:506-12. 19. Pan TS, King MA, Devries DJ, Ljungberg M. Segmentation of the body and lungs from Compton scatter and photopeak window data in SPECT: a Monte-Carlo investigation. IEEE Trans Med Imaging 1996;15:13-24. 20. Morozumi T, NAkajima M, Ogawa K, Yuta S. Attenuation correction method using information of attenuation distribution for single photon CT. Med Imaging Technol 1984;2:20-8. 21. International Commission on Radiation Units and Measurements. Tissue substitutes in radiation dosimetry and measurement, report 44. Bethesda (MD): International Commission on Radiation Units and Measurements; 1989. 22. Gullberg GT, Malko JA, Eisner RL. Boundary determination methods for attenuation correction in single photon emission computed tomography. In: Esser PD, editor. Emission computed tomography: current trends. New York: The Society of Nuclear Medicine; 1983. p. 33-53. 23. Murase K, Itoh H, Mogami H, Ishine M, Kawamura M, Iio A, et al. A comparative study of attenuation correction algorithms in single photon emission computed tomography (SPECT). Eur J Nucl Med 1987;13:55-62. 24. Soares EJ, King MA, Gtick SJ, Villegas BJ. Calculation of lung heart ratios for single photon emission computed tomography. IEEE Trans Nucl Sci 1996;NS-43:1995-9. 25. King MA, Tsui BMW, Pan TS, Glick SJ, Soares EJ. Attenuation compensation for cardiac emission computed tomographic imaging: part 2. Attenuation compensation algorithms. J Nucl Cardiol 1996;3:55-63.